Repair B0a baseline inventory invariants

[yingyue2030699]

Work package

B0a — baseline inventory logic repair.

Rebaselined with explicit user approval onto main@31f32e372819885a1c347598c401f3af74bf2de6 after the original kickoff baseline had advanced through reviewed Bfr documentation merges.

Primary claim

• Measure package-branch linearity from the origin/main fork point, so completed package merge commits already on main do not fail the inventory while merge commits introduced on an unmerged package branch remain rejected.
• Pin the historical WP0 ownership audit to the reviewed e9af3ddad494fc073040ee82bdf07944b9fee8cf..f8e76ea5bb444ba447a5ae9178a309545f2533ba range, keeping it fail-closed without admitting later packages as WP0 paths.
• Preserve generic_vs_cached_regular_median <= TBD as explicitly pending and reject any numeric substitution, including the superseded 1.10.

Evidence

• python3 -m py_compile scripts/inventory_unified_loop_baseline.py
• python3 scripts/inventory_unified_loop_baseline.py --check --json → errors: [], status: ok
• python3 -m unittest -v tests.test_unified_loop_baseline_inventory → 15 tests passed
• exact-head no-fast-forward current-main synthesis → passed at synthetic merge 29da3d692a06ad179cdb3d8a7a330896fe6bf871
• git diff 31f32e372819885a1c347598c401f3af74bf2de6..fda7a9957213d5763032c0aa253df20598b30a8c --check

No production, CUDA, workflow, fixture, route, or decision-status changes. Scientific review: N/A.

PR 186 disposition

This PR supersedes PR 186. PR 186 will be closed only after B0a merges and only on a later explicit user instruction; this PR does not merge, close, or retarget PR 186.